@1ay1/agenttyVerified Maintainer

agentty

Terminal AI assistants choke on long sessions — they repaint the whole transcript every frame and stall for seconds on b

Demo

Terminal preview

agentty

Overview

AI pair programming in the terminal with instant startup and support for various models

AI pair programming in your terminal — one static binary, sub-ms startup, any model

From the maintainer

Maintainer profile

agentty

Terminal AI assistants choke on long sessions — they repaint the whole transcript every frame and stall for seconds on b

Maintainer Website

AI pair programming in the terminal with instant startup and support for various models

Origin story

Terminal AI assistants choke on long sessions — they repaint the whole transcript every frame and stall for seconds on big threads. agentty makes an agentic coding CLI that stays instant no matter howlong you talk, on your own accounts and models.

First workflow

Point it at a real repo: "find bugs in X and fix them." Watch it map the code, fan out parallel explorers, edit, build, test — then measure instead of guess and commit. Investigate → verify → ship, all local.

Design choice

Rendering is O(viewport), not O(history) — enforced, not hoped. Settled turns are sealed once and handed to the renderer zero-copy, so a 2149-message / 12.6MB thread paints in 0.9ms with the full streaming typewriter animation at ~29µs/frame (0.17% of a 60fps budget). Fast is a structural invariant, guarded by perf tests.

Roadmap

Swap the JSON tokenizer to cut thread-open ~5× (the lexer is the bottleneck: 52ms to scan 12MB). Keep hardening the multi-provider / airgap / MCP surface. North star: always instant, always polished, your hardware.